Table 2. Single-structure evaluation: goodness of fit for the MTM and for SEM including a genomic $\left({\Lambda }_{\widehat{\mathit{U}}\mathit{*}}\right)$ or residual $\left({\mathbf{\Lambda }}_{\widehat{\mathit{E}}}\right)$ trait structure denoted by the BN algorithm it originates from.

BN Giving ${\Lambda }_{\widehat{\mathit{U}}\mathit{*}}$	BN Giving ${\mathbf{\Lambda }}_{\widehat{\mathit{E}}}$	DICa	pDb	logLc	No. Connectionsd	No. Nonnull Parameterse
Dent
—	GS 3	−87.0	+61.6	+66.0	10 + 6 = 16	15 + 13 = 28
—	GS 1, 2, 4	−78.5	+65.7	+72.1	10 + 5 = 15	15 + 12 = 27
—	TABU 1, 2	−62.7	+45.1	+23.3	10 + 6 = 16	15 + 15 = 30
TABU 1	—	−0.7	−12.6	+12.1	8 + 10 = 18	15 + 15 = 30
—	—	0	1024.9f	0	20	30
TABU 2	—	+1.0	−14.1	−7.3	9 + 10 = 19	15 + 15 = 30
GS 1, 2, 3, 4	—	+57.0	−16.9	−32.4	7 + 10 = 17	15 + 15 = 30
Flint
—	TABU 1	−124.1	+64.0	+105.2	10 + 5 = 15	15 + 13 = 28
—	GS 1, 2, 3, 4	−122.3	+63.3	+126.3	10 + 5 = 15	15 + 13 = 28
—	TABU 2	−117.7	+61.0	+131.1	10 + 6 = 16	15 + 14 = 29
—	—	0	1086.0f	0	20	30
TABU 1, 2	—	+35.8	+23.1	−3.2	7 + 10 = 17	14 + 15 = 29
GS1, 2, 3, 4	—	+72.1	+4.2	−57.1	6 + 10 = 16	12 + 15 = 27

For notation of BN algorithms see Material and Methods, Learning genomic and residual BN.

^aDeviance information criterion: DIC of SEM minus DIC of MTM.

^bEffective number of parameters: p_D of SEM minus p_D of MTM.

^cLogarithm of Bayesian marginal likelihood: logL of SEM minus logL of MTM.

^dSum of connections in the networks: genomic plus residual.

^eSum of nonnull parameters in the covariance matrices: genomic plus residual.

^fAbsolute value of p_D.